Community Membership and Reciprocity in Lending: Evidence from Informal Markets - Wharton ...

Community Membership and Reciprocity in Lending: Evidence
                  from Informal Markets?
                                          Rimmy E. Tomy∗
                          The University of Chicago, Booth School of Business

                                   Regina Wittenberg-Moerman
                                    University of Southern California

                                            January 25, 2021

                                              Abstract

We study how wholesalers extend trade credit to retailers in economies where formal mar-
ket institutions, such as financial reporting systems, auditing, and courts, are nonexistent or
function poorly. Using the setting of a large market in the northeastern part of India, we find
that community membership plays a strong role in the access to trade credit. Wholesalers are
more likely to provide trade credit and offer less restrictive credit terms to within-community
retailers, are more lenient when these retailers default, and are less likely to experience de-
faults from them. We show that this cooperation between same-community wholesalers and
retailers is achieved through a reciprocity mechanism, which provides insurance against in-
come shocks.

JEL Classification: D82, G21, G28, O10, O16, O17, Z10, Z13
Keywords: Trade Credit, Informal Economies, Lending, Reciprocity, India, Iewduh, Com-
munity Enforcement, Asymmetric Information

  ?
     For helpful discussions and comments, we are grateful to Ray Ball, Marianne Bertrand, Ron Burt,
Pradeep Chintagunta, Anna Costello, Doug Diamond, Jung Koo Kang, Christian Leuz, Kaivan Munshi,
Shailesh Nagar, Tiplut Nongbri, Rohini Pande, Jane Risen, Raghuram Rajan, Abbie Smith, Amir Sufi,
Chad Syverson, George Wu, and Luigi Zingales. We thank seminar participants at the University of Toronto,
Tulane University, and USC for their helpful comments and suggestions. We are grateful to Shailesh Nagar
and NRMC for help with the data collection. Rimmy E. Tomy gratefully acknowledges the support of the
Fama Miller Center and IGM at the University of Chicago Booth School of Business.
   ∗
     Corresponding author
     Email addresses: Rimmy.Tomy@chicagobooth.edu (Rimmy E. Tomy), reginaw@marshall.usc.edu
(Regina Wittenberg-Moerman)

1. Introduction

      A large body of literature examines the economic factors that affect private lenders’
decisions to extend credit to borrowers (Berger et al., 2005; Diamond, 1984; Fama, 1985;
Jappelli et al., 2005; Petersen & Rajan, 2002). Yet, we know little about how borrower risk
is assessed and credit is extended in markets where auditing and financial systems function
poorly and there is limited recourse to the legal system in case of nonrepayment. In this
study, we extend prior research by exploring factors that influence wholesalers’ decisions to
extend trade credit to retailers in an informal sector of a developing economy and focus on the
role of community membership in facilitating the flow of trade credit. Our focus is motivated
by prior research that highlights the importance of in-group lending in developing economies
(Besley & Coate, 1995; Kinnan & Townsend, 2012; Udry, 1994; McMillan & Woodruff, 1999).
      The informal sector accounts for about 30% of GDP and 70% of employment in emerging
market and developing economies (World Bank, 2019).1 This sector is broadly defined by
economic activity that is not de facto or de jure regulated or protected by the state. Even
though these activities may be carried out within the formal reach of the law, regulations
are not enforced because it is too costly or is not feasible to do so (OECD/ILO, 2019). The
informal sector employs a large workforce and has low levels of development and productivity.
Businesses in this sector have limited access to formal sources of credit from banks and capital
markets, and therefore rely heavily on trade credit. An understanding of the factors that
can ease access to trade credit in the informal sector can potentially improve trade and
investment in this sector, aiding its long-term growth (Fisman & Love, 2003; Petersen &
Rajan, 1997; Rajan & Zingales, 1998).
      The limited insight into the informal economy is driven by the lack of organized data
sources and the difficulty in accessing this largely unorganized sector. We approach this
challenge by surveying wholesalers and retailers in a large organized marketplace, or bazaar,

  1
    Figure 1, sourced from OECD/ILO (2019), shows the distribution of informal employment, which dom-
inates in the Global South.

                                                 1

in the northeast of India, in the state of Meghalaya. The marketplace, called Iewduh, is
the epicenter of trade in the region (Figure 2 shows the geographic location of Iewduh). We
select this market as the setting for our study for several reasons. First, retailers in Iewduh
rely on trade credit from wholesalers. Second, this market is characterized by the presence
of multiple products and ethnic groups, allowing us to take advantage of the rich variation
in the data. The market includes traders from several communities, defined as those sharing
a common language, culture, and place of migration. Third, the market is a microcosmic
representation of a larger sector of small business enterprises across Africa, Asia, and Latin
America, thereby mitigating generalizability concerns. Fourth, the presence of wholesalers
and retailers in a relatively constrained area reduces the cost of conducting a survey. Finally,
focusing on a single market allows us to communicate directly with the traders who comprise
the observations in our sample and to understand the social and economic constraints within
which they operate and which affect their credit and business decisions.
   We interview a probability sample of 503 traders operating in the market, of which
146 are wholesalers and the remaining are retailers. We ask the traders about their links
to other traders, resulting in a data set with information on 1,230 wholesaler-retailer links
(relationships). We collect data on whether trade credit is provided, the terms of trade credit,
wholesalers’ responses to retailer defaults, the parties with whom wholesalers communicate
about retailers, the information wholesalers collect directly from retailers, and the actions
they take to collect it.
   The observed amounts and terms of trade credit reflect both the supply and demand for
credit. However, our analyses should primarily capture the effect of community membership
on the supply of credit because the lack of formal sources of financing makes retailers credit
constrained and heavily reliant on trade credit from wholesalers. That is, all retailers demand
credit but access to it is somehow rationed by wholesalers. Relying on the exogenous nature
of community membership, which is determined by birth, and a within-wholesaler estimation,
we find that wholesalers are more likely to lend to retailers from within their community and

                                               2

are less likely to experience defaults by these retailers. Specifically, wholesalers are 11.5%
more likely to provide trade credit to retailers from their community relative to retailers from
other communities. Conditional on providing credit, wholesalers provide 8% more credit to
retailers within their community, and face 13.6% less default from these retailers. We also
find that wholesalers are more lenient towards defaulting retailers from their community:
they are less likely to repossess goods, deny future credit, or take other actions against
them. That wholesalers are more lenient towards same-community retailers in case of default
might imply that retailers should repay outside-community wholesalers before repaying same-
community wholesalers, resulting in higher within-community defaults. Our opposing finding
that wholesalers face lower defaults from same-community retailers suggests the presence of
a mechanism that incentivizes same-community wholesalers and retailers to cooperate.
   We contend that a norm of reciprocity explains such wholesaler-retailer cooperation.
Reciprocity can be indirect: a wholesaler who is more generous towards a retailer may derive
benefits from other community members who are aware of their generosity (Ferrara, 2003;
Nowak & Sigmund, 2005). Because credit allocation within communities insures against
income fluctuations (Besley & Coate, 1995; Kinnan & Townsend, 2012; Udry, 1994), we
conjecture that reciprocity within the community is a response to income instability and a
lack of formal sources of insurance, and is sustained by members’ beliefs that adherence to
the norm will lead to community support in times of need. In case of shocks to income,
traders rely on community members for help and, in return for this insurance, assist other
community members even when it is costly. The cost to wholesalers arises from having to
obtain additional resources to support larger amounts of trade credit to same-community
retailers, while the cost to retailers relates to them not exploiting the leniency of same-
community wholesalers.
   To provide evidence in support of the reciprocity mechanism, we examine whether whole-
salers who are more likely to benefit from the norm of reciprocity are also more likely to
follow it. We find that wholesalers who do not have a stable source of income (e.g., rental

                                               3

income, family members with a stable job), and therefore depend on their community in
times of economic distress, provide 17% more credit to within-community retailers. Simi-
larly, wholesalers who have low levels of education and therefore fewer outside options are
also more likely to rely on community members in case of income shocks. These wholesalers
provide 14% more credit and offer relatively longer repayment times to within-community
retailers. In addition, wholesalers who also have some retail business are direct beneficiaries
of the reciprocity norm because they not only provide but also receive trade credit. Such
wholesalers are 23.5% more likely to provide trade credit, and conditional on providing credit,
lend 42% more to retailers from their community than to those from other communities.
   Indirect reciprocity requires within-community information flows because community
members cannot reciprocate unless they are aware that a member behaves in a manner con-
sistent with this norm (Nowak & Sigmund, 2005). Experimental research also highlights the
importance of an information channel in sustaining indirect reciprocity, as it allows members
to develop a reputation for reciprocity (Bolton et al., 2005; Engelmann & Fischbacher, 2009;
Seinen & Schram, 2006; Wedekind & Milinski, 2000). We contend that the community is
more likely to form accurate beliefs about members’ reciprocal behavior when these members
have greater ties to the community, which facilitates information transmission. Consistent
with this argument, we find that preferential within-community lending is stronger among
wholesalers whose family members are all married within the community, and among those
who meet more often with their community members for social events. These wholesalers
provide 22%–38% more credit and offer more beneficial credit terms to retailers from their
communities. Conversely, wholesalers who have significant business contacts outside their
community are less likely to develop a reputation for within-community reciprocity relative
to wholesalers whose business contacts are all within their community. Wholesalers with sig-
nificant business contacts outside their community are also less likely to be dependent on the
community. We find that these wholesalers are 29% less likely to provide credit, provide 18%
lower amounts of credit, and offer shorter repayment times. Overall, these analyses based

                                              4

on wholesalers’ incentives to reciprocate and their ties to the community are consistent with
a reciprocity mechanism contributing to the provision of trade credit.
   To supplement the wholesaler-based analyses, we explore characteristics of retailers that
are associated with their incentives to reciprocate, such as having no stable sources of income
or a low education. We do not find that these characteristics explain retailers’ propensity
to default on trade credit from within-community wholesalers, potentially because, relative
to wholesalers, retailers tend to be smaller and poorer and therefore the majority rely on
their community for insurance against shocks to income. With respect to how the commu-
nity assesses retailers’ reputations for reciprocity, we do not find that wholesalers publicize
instances of same-community retailer defaults, possibly because doing so may be seen in
bad taste by other community members who could believe that the retailer has fallen on
hard times. Thus, the greater leniency towards same-community retailers who default may
enhance a wholesaler’s reputation as one who helps the community. However, in additional
analyses we find that, to collect information about same-community retailers, wholesalers
are more likely to speak with other wholesalers from their community, which should help the
community form beliefs about retailers’ incentives to reciprocate.
   To further support the reciprocity mechanism, we focus on the lending behavior of immi-
grant communities, to which the majority of traders in our sample belong. Newer immigrant
communities, which we identify based on community growth, are more likely to face large
income fluctuations, necessitating the reliance on other community members to smooth in-
come. Traders in small communities are also more likely to have stronger community ties
(Munshi, 2011), facilitating the formation of accurate beliefs about other traders’ incentives
to reciprocate. Consistent with the norm of reciprocity, we find that wholesalers’ preferential
lending is stronger in small, high-growth immigrant communities and is also present, albeit
weaker, in small, low- and medium-growth immigrant communities. Moreover, retailers in
small immigrant communities are less likely to default on trade credit from same-community
wholesalers. Thus, similar to wholesalers, retailers who are more dependent on their com-

                                              5

munity and who have a greater ability to establish a reputation for collaboration behave
according to the norm of reciprocity.
   Finally, we conduct additional interviews with twenty traders in our sample following the
COVID-19 related lockdown and the associated income shock to gauge their beliefs about
community support. Many traders responded that they felt a responsibility to help their
community members and believed that community members would help them in return.
Traders’ responses corroborate our proposition that a reciprocity mechanism within the
community helps members deal with income shocks.
   We next address potential alternative explanations for our findings of preferential within-
group lending and lower retailer defaults. Prior literature suggests that taste-based discrim-
ination may explain preferential in-group lending (Becker, 1957; Blanchflower et al., 2003;
Ladd, 1998). However, such discrimination presumes that lenders would be willing to incur
additional costs, such as higher default rates, for favoring in-group members (Haselmann
et al., 2018; Fisman et al., 2020). Our finding of a lower likelihood of default for within-
community retailers suggests that a taste-based discrimination mechanism is unlikely to drive
our findings. It is also possible that statistical discrimination, which assumes that lenders
are unable to form accurate beliefs about the creditworthiness of out-group members due
to inferior information, explains preferential within-community lending. Our evidence that
wholesalers have extensive and long-term lending relationships with and therefore substantial
knowledge about members of outside communities suggests that statistical discrimination is
also unlikely to explain our findings. Furthermore, our findings that wholesalers who are
more likely to face income fluctuations lend more and offer better terms to their community
retailers are inconsistent with taste-based and statistical discrimination.
   Our findings could also be attributed to the asymmetric information mechanism (De Meza
& Webb, 1987; Stiglitz & Weiss, 1981). Under this mechanism, the preferential within-
community lending and the lower within-community default rates that we document could
be driven by both better ex ante screening of retailers and more efficient ex post enforce-

                                              6

ment by the community, as within-community information sharing facilitates these channels.
(Fisman et al., 2017; Greif, 1993; McMillan & Woodruff, 1999). Two sets of findings reveal
that an asymmetric information mechanism is unlikely to be the primary explanation for our
findings. First, better ex ante screening due to a more precise evaluation of the creditworthi-
ness of within-community retailers should result in a higher variation in the terms of trade
credit for within-community retailers relative to retailers from other communities (Cornell
& Welch, 1996; Fisman et al., 2017). In contrast, we find a smaller variation in the amount
of trade credit for lending within the community, which rather suggests that wholesalers are
being less discerning within the community. Second, our finding that wholesalers are more
lenient towards same-community retailers who default indicates that more efficient ex post
enforcement within the community does not drive the preferential lending we observe. This
argument is further supported by our finding that wholesalers do not spread information
about defaults. Although we find that wholesalers are more likely to speak about within-
community retailers with wholesalers from their community, our findings are overall more
consistent with within-community information flows allowing the community to assess retail-
ers’ incentives to reciprocate rather then with ex post enforcement. Moreover, our findings
of stronger preferential within-community lending by wholesalers who are more dependent
on their community cannot be explained by ex ante screening or ex post enforcement.
   Finally, we conduct additional analyses to mitigate concerns that our findings could be
attributed to personal relationships between traders, which are likely to be more prevalent for
within-community traders. These relationships may facilitate access to credit due to lower
information asymmetry between connected traders, or wholesalers’ favoritism toward traders
with whom they have personal relationships (Engelberg et al., 2012; Haselmann et al., 2018).
Our findings that wholesalers are not more likely to lend to same-community retailers the
first time they transact or to provide them with credit after a fewer number of cash transac-
tions, as would be the case if personal relationships were driving credit availability, indicate
that preferential within-community lending is unlikely to be driven by wholesaler-retailer

                                               7

personal relationships. Furthermore, lending based on personal relationships presumes that
wholesalers have substantial information about same-community retailers from prior social
interactions, reducing the need for information collection. We study wholesalers’ informa-
tion collection decisions, including information on the location of retailers’ shops and the
length of time retailers’ businesses have existed. We find that wholesalers do not collect less
information from same-community retailers relative to other retailers, as lending based on
personal relationships would imply. Moreover, wholesalers exert as much effort (measured
as visiting retailers’ shops or calling them) to actively collect information about retailers in
their community as about those outside their community. These results further reinforce our
inference that personal relationships cannot be the main driver of our findings.
       Our work contributes to the understanding of credit flow in the informal sector. Prior
studies document that because of weak enforcement and the difficulty in assessing borrower
risk, businesses in this sector have limited access to formal sources of credit, such as bank
loans (Hoff & Stiglitz, 1990). Our study highlights the role of community membership
in facilitating access to credit in the informal sector. In this respect, our findings also
complement recent work that studies how the introduction of formal financial institutions,
such as microfinance or development programs, influences credit availability and growth
in developing economies (Banerjee et al., 2013, 2015). While these institutions are often
beneficial, they also crowd out both informal lending relationships and social interactions
unrelated to lending (Banerjee et al., 2018; Heß et al., 2020). Our evidence of the role
of community in providing access to trade credit and thus insuring against income shocks
underline the risks of eroding informal economic structures.2
       Our work further adds to studies that explore how group membership and social net-
works alleviate informational frictions in lending (Engelberg et al., 2012; Fisman et al.,
2017; McMillan & Woodruff, 1999). We extend these studies by showing the importance of

   2
      Rajan (2019) highlights the important role that the community plays even in developed economies, where
it fills the gaps left by formal government and market systems.

                                                     8

a reciprocity mechanism in the provision of trade credit in informal economies. Our work
also connects studies of in-group lending to those that explore the reciprocity that arises in
informal economies to deal with income fluctuations and the lack of formal insurance (Besley
& Coate, 1995; Kinnan & Townsend, 2012; Udry, 1994). We document that the wholesalers
who are more likely to rely on communal reciprocity in the case of a shock to their income
are also the ones who assist community members by providing them with preferential access
to trade credit.
   Importantly, our study is distinct from work that explores the effect of cultural similarities
on bank lending. Fisman et al. (2017) use data from a large bank in India and conclude that
cultural proximity between loan officers and borrowers serves to reduce informational frictions
in lending. While Fisman et al. (2017) suggest that cultural proximity mitigates these
information frictions due to shared commonalities, such as codes, language, and religion, we
propose that reciprocity incentivizes same-community retailers and wholesalers to cooperate.
Furthermore, Fisman et al. (2017) explore cultural proximity through the prism of a bilateral
relationship between the loan officer and the borrower, but the mechanism of reciprocity
in our setting extends beyond bilateral relationships and is likely to encompass the entire
community. Overall, we demonstrate that, as opposed to a simple dyadic exchange, the flow
of credit in the informal sector is embedded in a larger social structure.

2. Survey methodology

   Our objective is to study the flow of credit in informal economies and to understand the
frictions that prevent access to credit. Because businesses that operate in informal economies
typically do not borrow from formal sources of capital and do not report financial data, we
rely on the survey method to collect data from traders in a large bazaar in India. The survey
includes questions about operations of the business with a focus on lending and access to
credit. In this section, we describe our data and the data-collection process.

                                               9

2.1. Feasibility study

       We conducted a feasibility study to understand the features of the market and the via-
bility of a study there.3 The feasibility study involved in-depth interviews with twenty-two
traders in the market and limited data-collection from forty-nine additional traders. The
study revealed that the use of trade credit is prevalent in the market. Wholesalers often
need to provide trade credit to distribute their goods because many retailers cannot afford
to pay cash upfront for larger amounts of goods. On the other hand, retailers tend not to
provide credit to their customers because of the difficulty in tracking them. Furthermore,
trade credit is an important source of financing for retailers because they tend not to access
loans from the formal banking system. We found that several factors inhibit access to formal
credit institutions. These include a lack of understanding of the loan application process,
low levels of education, lack of trust in formal institutions, an inability to provide land or
other assets as collateral, discomfort with being indebted, and religious reasons for not taking
loans with explicit interest payments.
       We also found considerable product-level variation in the terms of credit provided. For
example businesses that provide services, such as salons or electrical goods repair shops, tend
not to use trade credit. Traders of highly perishable products, such as raw meat, also do not
use trade credit. On the other hand, retailers of general provisions typically receive three
to four weeks of credit, whereas retailers of metal goods, such as knives and agricultural
implements, receive, on average, only four to five days of credit.
       The feasibility study also revealed the typical terms of trade credit in the market. Retail-
ers typically do not receive credit in their first transaction with a wholesaler. Wholesalers
start a new credit relationship with a retailer by extending a small amount of credit. Once
the retailer establishes a reputation for paying back on time, the wholesaler increases the

   3
    The feasibility study was conducted in August 2018. In July 2019, we carried out a listing exercise to
create a list of all shops in the market. Piloting of the survey questionnaire was conducted in October 2019,
and the main survey ran from December 2019 to February 2020.

                                                     10

credit limit. Wholesalers do not price discriminate based on credit provision. That is, re-
tailers do not receive a discounted price if they pay in cash versus if they take credit. Also,
there is no early payment discount and no interest charged for delayed payment. Thus, the
variation in credit terms comes primarily from the amount of credit provided and stipulated
repayment time. Some wholesalers also provide trade credit with no stipulated repayment
time. In such cases, the retailer repays the wholesaler whenever they transact next.
   Interestingly, wholesalers have little recourse in case of default. Many wholesalers, who
tend to be immigrants to the region, can neither repossess goods, nor threaten the use of
force because of the multiple ethnic groups operating in the market and the possibility of
ethnic conflict such actions can incite. Also, because courts are inefficient, entering into
litigation is generally not an option. However, cases where the retailer never repays are rare.
Given that nonrepayments are rare, default events in our study are defined as instances
where the retailer does not repay within the stipulated time.
   Finally, the feasibility study helped us define the communities operating in the market.
These are primarily based on ethnicity, which we interpret as sharing a language, culture and
place of migration, with the exception of the Muslim community, which is based on religion.
We define communities in this manner because these are the groups with which respondents
identify when asked to state their ethnicity. Specifically, the groups are the Khasi, Jaintia,
Marwari, Bengali, Bihari, Nepali, Punjabi, Assamese, and Muslim communities. Khasis are
the dominant ethnic group and are indigenous to the region, whereas the remaining groups
are largely migrants who came into the state from other parts of the country or from other
countries. Although Muslims could also belong to another group (for example, Bengali),
they self-identify as belonging to the Muslim community. Self-identification is important in
our setting because our objective is to discern the social network traders belong to and the
information they can access. Therefore, although defined by religion, Muslims in the region
can also be considered to be an ethnic group. We did not ask respondents about their caste
because, being traders, they belong to similar castes.

                                              11

2.2. Sampling

      To select a representative sample of the market, we created a list of all businesses in
the market. This exercise resulted in a list of 5,254 shops with information on the location,
product category, and gender of the trader. Because the listing exercise was based on obser-
vation rather then direct interaction with the trader, we were not able to collect additional
auxiliary information, such as whether the trader is a wholesaler or retailer, or the ethnicity
of the trader. Several shops were unoccupied or could not be classified into distinct product
categories (for example, warehouses) and are eliminated from the list, which left us with
4,437 shops.
      Because our objective is to explore access to credit, we retain only product categories
where trade credit is prevalent based on our findings from the feasibility study. After elimi-
nating shops in product categories where trade credit is not commonly used (as described in
Section 2.1), we retain 2,537 shops. Lastly, to ensure an adequate number of shops within
each product group, we further restrict shops to product groups with at least 150 shops.
This requirement results in a target population of 1,877 shops and consists of the follow-
ing product categories: general stores, footwear, household appliances, textiles, tobacco and
betel.
      We draw a probability sample from the target population by using a stratified random
sampling approach, where the strata are defined as product groups. We randomly select
traders (shops) within each stratum.4 Probability sampling allows us to pick a representative
sample so that inferences drawn from the sample can be generalized to the population. Given
that the terms of trade credit are related to product type, the variance of our outcome
variables are likely to be higher across product groups than within them. Therefore, the use
of stratified random sampling increases the precision of our estimates (Frankel, 2010). Our
sampling unit is a trader who is the sole proprietor of a shop (each trader is associated with

  4
    We used a disproportionate stratified random sampling approach where we oversampled from strata with
low response rates. We adjust our sample weights for the oversampling.

                                                  12

only one shop). A trader can be a wholesaler, a retailer, or both. In cases where traders state
that they engage in both wholesale and retail business, we require at least 70% of the total
sales to come from the wholesale business in order to classify the traders as wholesalers. We
impose these criteria because we are interested in classifying traders primarily as providers
or recipients of trade credit. Wholesalers, by definition, are fewer in number. Therefore,
we oversample wholesalers by using disproportionate sampling based on screening.5 When
performing the empirical analysis, we adjust for the oversampling and nonresponses in our
sample weights (Frankel, 2010).
       Table 1, Panel A shows the distribution of the population, sample, and response rates by
strata. Column (1) shows the number of shops by product category in the target population
and column (2) shows the percent distribution of the shops. Textile shops comprise the
largest share with 38% of all shops in the population falling in this group. General stores
comprise the next highest share at 17%, with the remaining categories accounting for between
9% to 13% of the total. We select a sample size of 612 traders, which is approximately one-
third of the target population of 1,877. Columns (3) and (4) of Table 1, Panel A show
the number of shops and percent distribution of shops in the survey sample. The sample
distribution corresponds to the population distribution. Finally, columns (5) and (6) of the
table show the number of responses and response rates by product category. The overall
response rate is 82% and varies by product type, ranging from 70% for household appliances
to 91% for tobacco.
       We interviewed thirteen wholesalers and forty-five retailers in a pilot study and asked
them for information about their top-ten connections, based on the percent of sales (pur-
chases) in the previous year for wholesalers (retailers). On average, wholesalers had connec-
tions to 2.08 retailers and retailers to 1.84 wholesalers. The response rate in the pilot survey

   5
    To achieve a larger sample of wholesalers, after ascertaining whether the trader is a wholesaler or
retailer, we oversample wholesalers by continuing the full interview at different rates for wholesalers and
retailers (Piazza, 2010). We were able to estimate the proportion of wholesalers in the market, by product,
by randomly selecting traders to interview.

                                                    13

was high at 94%. Therefore, a sample size of 612 traders was expected to yield data on over
1,000 wholesaler-retailer links.6 The link between a wholesaler and retailer is the unit of
observation in our empirical analyses. We conduct a power analysis to assess the adequacy
of our sample size (Land & Zheng, 2010). Figure 3 shows a plot of power (the probability of
avoiding a Type II error) by sample size for values of the multiple-partial correlation between
the predictor variables and outcome ranging from 0.05 to 0.2, at a 0.05 significance level.
As can be seen from the plot, at 80% power and with a sample of 1,000 wholesaler-retailer
links, we should be able to detect a partial correlation of 0.08 and above.
    Our final sample includes survey responses from 503 traders. Of these, 146 are wholesalers
and the remaining are retailers. Combined, they provide data on 1,230 wholesaler-retailer
links.7 Given that the average number of links identified in the pilot survey was approxi-
mately two and the maximum number of links was five, we asked traders for their top five
links in the main survey.8 A wholesaler in our final sample has, on average, links to 3.40
retailers, whereas a retailer has links to 2.06 wholesalers. Traders have links both within the
market and outside. Based on the locations of wholesalers’ and retailers’ shops, about 50%
of the reported links are within the surveyed market.

2.3. Survey instrument and data

    The survey instrument consists of three main sections. The first section is relatively short
and asks for identifying information, such as the shop name, the trader’s name, and whether
the trader is a wholesaler or retailer. The second section is the core of the survey and contains

    6
      The calculation is as follows: 612×94%×2. In the final survey, our response rate was only 82%, however,
traders reported a greater number of links.
    7
      We manually combine data from the responses of wholesalers and retailers using the trader’s name, shop
name, shop location, telephone number, gender, and ethnicity. That is, if retailers mention a wholesaler we
interviewed (and vice-versa), we form a connection between the two traders if none existed before and assign
to it wholesaler (retailer) characteristics obtained from the respective wholesaler (retailer) interviews. All of
the links in our sample are one-sided (i.e., we only interview one party in the wholesaler-retailer link). One
reason for this is that traders have extensive links outside the market.
    8
      As we report in Panel B of Table 1, a link on average accounts for 20.72% of the sales (purchases) for a
responding wholesaler (retailer), further suggesting that limiting our survey to the top-five links is unlikely
to bias our results.

                                                       14

six main questions with several subparts. For each of the top-five wholesalers or retailers we
ask traders to identify, we require details related to the provision and terms of trade credit.
This includes whether trade credit is provided to the retailer, if credit was provided the first
time that the wholesaler and retailer traded, the number of cash transactions before the
wholesaler provided credit, the amount of trade credit provided, and the repayment time.
We also ask traders about the incidence of defaults. Wherever relevant, we ask traders to
provide a range of estimates. Besides questions related to credit access, we ask traders about
the types of information they collect and the actions they take to collect that information, as
well as about trader-specific characteristics, such as the number of years that the wholesaler
and retailer have been in business, the community and gender of the other party, and the
share of sales. Finally, part three of the questionnaire includes questions related to the
responding trader’s demographic characteristics.
      We took several measures to ensure the accuracy of responses. Prior research has found
greater levels of missing data and less detailed answers for items that appear later in the
questionnaire (Krosnick & Presser, 2010). Therefore, questions related to demography were
asked towards the end of the questionnaire to minimize fatigue effects because traders would
need to exert relatively less effort to answer basic demographic questions about themselves.
Furthermore, our survey team consisted of twelve enumerators who were local to the region
and were well versed in the language and customs. The enumerators were carefully selected
from a pool of applicants who were mainly graduate students from local universities. A
three-day training session was conducted for the enumerators. All interviews were conducted
face-to-face and the data were recorded electronically in a survey application. The software
also recorded the location coordinates of the interviews and the time taken to complete the
questionnaire. The survey managers conducted 20% spot checks and 10% back checks.9 As
an additional check, with the traders’ permission, we asked enumerators to take photographs

  9
    A spot check is when survey managers observe the enumerators conducting the interviews. In a back
check, survey managers revisit the respondent and ask selected questions again.

                                                 15

of themselves with the trader in the place where the interviews were conducted.
       Table 1, Panel B presents descriptive statistics for the variables used in our primary
analyses for our sample of 1,230 wholesaler-retailer links. Of these links, 33% are between
traders belonging to the same community, and 57% are between traders of the same gender.
A link, on average, accounts for 20.72% of sales for the responding trader. The mean
relationship length between the wholesaler and retailer is 7.8 years. Geographically, 53.3%
of the traders in our sample are located close to their trading partners, which we define as
a radius of less than or equal to three kilometers.10
       The use of trade credit is prevalent in our product categories, with 893 (72.6%) of the
trading links having some amount of trade credit. For the remaining 27.4%, we asked
wholesalers (retailers) their reasons for not providing (receiving) trade credit. The responses
included prior nonrepayment of credit, the distance from the retailers’ shop being too great,
a lack of trust, the retailer not comprising a significant portion of sales, the wholesaler being
too small and having a policy of not giving credit to anyone, the retailer being new to the
business, the wholesalers’ parents (in the case of an inherited business) never giving credit,
and the retailer visiting the wholesaler’s shop infrequently.
       Of the 893 cases where credit is provided, only in 30.7% instances is trade credit provided
on the first transaction between the wholesaler and retailer. On average, there are 5.35
cash transactions before credit is provided and the mean amount of credit equals 40.45% of
sales. With respect to the cost of trade credit, as discussed in Section 2.1, the cost varies
primarily along the repayment-time dimension; 32.4% of the links are associated with a short
repayment time, defined as fifteen days or less. However, in 20.7% of the cases the repayment
time is not fixed. That is, the retailer can repay the wholesaler whenever they transact next.
Finally, default occurs in 18% of our sample, where default is defined as the retailer not
repaying within the stipulated time.

  10
    We define distance based on this indicator variable because the market is densely packed and it is
therefore difficult to accurately measure distance between shops, especially for shops that are located nearby.

                                                      16

We have data on wholesaler characteristics for about 40% of the sample (for 496 wholesaler-
retailer links from the 146 wholesalers we interviewed). Wholesalers we interviewed have been
in business for an average of seventeen years and rely primarily on income from the business
(76.2% have no other stable sources of income). As far as education is concerned, 47.3% do
not have a high school degree. Because traders are reluctant to share information on the
size of their businesses, we asked for responses on a relative scale from 0 (low) to 10 (high).
The average wholesaler reports a size of 5.8. We also collect information on size using an
absolute, but coarser scale (untabulated): a majority (59.5%) of wholesalers report total
asset value in the range of INR 100,000–1,000,000, which translates to $1,359–$13,587 at an
exchange rate of 73.6 INR/USD.
   We also asked wholesalers for information to assess their ties to the community. First,
we asked wholesalers about their significant business ties, which are defined as contacts
who provide wholesalers with information related to the product area and market trends,
contacts who have helped them through a significant crisis event in their business career, or
their most valuable employee. Only 6.6% of wholesalers have significant business ties outside
their community. Second, in 96% of cases, all family members of the wholesaler are married
within the community. Third, 20.7% attend community social events at least once a week,
which we classify as high social event attendance. Furthermore, few family members are
involved in the business, ranging from 0 to 3 with an average of 0.5. Thirty-one percent of
wholesalers also engage in some retail business. Finally, 46.4% are entrepreneurs (i.e., have
not inherited a business) and the wholesaler is female in 15.5% of the cases.
   Finally, we collect information on the population of these communities from the 2001 and
2011 Censuses of India. We identify community population at the district level (East Khasi
Hills district) by using mother tongue and religion tables from the censuses. We classify
immigrant communities as large and small based on their share of the immigrant population
in the district. We define large immigrant communities as those that comprise 15% or
more of the immigrant population in the region. Small immigrant communities are further

                                              17

divided into high-growth (ten-year growth of greater than 100%), and low- and medium-
growth communities (all large immigrant communities have a low growth). As reported in
Table 1, Panel B, 29% of our sample links belong to large immigrant communities, 31.5%
to small low- and medium-growth immigrant communities, and 10.7% to small high-growth
immigrant communities.

3. Empirical specification and main results

3.1. Provision of trade credit and the incidence of default

   Our baseline empirical specification identifies the effect of community membership on the
provision and terms of trade credit. An advantage of our setting is that group membership
is exogenous and determined by birth. To control for unobserved wholesaler-specific charac-
teristics that can drive trade credit related decisions, we examine how the same wholesaler
differs in her lending behavior to retailers within versus outside their community. We are
able to operationalize a within-wholesaler design because wholesalers in our sample have
multiple retailer links that span across communities. Specifically, we estimate variations of
the following OLS model:

                    Yij = α + θSame Communityij + βk Xij + λi + eij ,                     (1)

where i indexes wholesalers and j retailers. The dependent variable Y represents trade
credit related outcomes, and indicates whether the wholesaler provides trade credit to the
retailer, the terms of credit provided, and the incidence of default; Same Community is an
indicator variable that equals 1 if the wholesaler and retailer belong to the same community
and zero otherwise; X is a vector of k control variables and includes link-specific charac-
teristics. Specifically, X includes the number of years that the wholesaler and retailer have
had a trading relation (Relationship Length), the share of total sales that flows through the
wholesaler-retailer link as a measure of link importance (Share of Sales), whether the whole-
saler and retailer are of the same gender (Same Gender ), and a measure of the geographic

                                             18

distance between the wholesaler’s and retailer’s shops (Distance (Close)). Wholesaler-specific
characteristics are captured by wholesaler fixed effects, λi : the inclusion of λi implements
the within-wholesaler estimation. Finally, e is the error term.
   The observed amounts and terms of trade credit reflect both its supply by wholesalers
as well as retailers’ demand for credit. However, our analyses rest on the assumption that
all retailers demand credit, but access to it is somehow rationed. Our interviews suggest
this assumption is reasonable because retailers tend not to rely on bank loans as a source
of financing, but rather on personal funds or loans from relatives and friends. As discussed
in Section 2.1, although retailers may have access to loans from banks, they tend to be
skeptical of the process and the cost of borrowing from banks. The lack of formal sources
of financing makes retailers heavily reliant on trade credit from wholesalers. Therefore, our
research design should allow us to primarily capture the effect of community membership on
the supply of credit.
   Table 2 presents results from the estimation of Equation 1. The dependent variable in
column (1) is an indicator for whether the wholesaler provides trade credit to the retailer.
Columns (2) and (3) indicate whether, conditional on providing credit, the wholesaler pro-
vided credit the first time she traded with the retailer, and the number of cash transactions
before credit was provided for the first time. Columns (4) to (6) provide additional details
related to the terms of trade credit: the dependent variable in column (4) is the amount
of trade credit provided as a percentage of sales in the last year, whereas columns (5) and
(6) provide measures of the repayment time. Finally, column (7) measures whether the
wholesaler experienced default in the last year.
   The results indicate that wholesalers are 11.5% more likely to provide trade credit to
retailers from their community relative to retailers from a different community. Further-
more, conditional on providing credit, wholesalers provide 8% more credit to retailers within
their community. These results indicate preferential lending to within-community retailers.
However, there is no evidence that wholesalers provide same-community retailers credit in

                                             19

the first transaction or after a fewer number of cash transactions. Community membership
also does not endow retailers with more beneficial repayment times. Finally, wholesalers are
13.6% less likely to experience defaults from same-community retailers relative to retailers
from other communities. In robustness analyses (untabulated), instead of wholesaler fixed
effects, we include product and wholesaler community fixed effects. Our results continue to
hold. Our results are also unchanged when we include enumerator fixed effects, to account
for any idiosyncratic effects of enumerators. Finally, our results continue to hold when we
control for whether wholesalers collect information about the retailer before providing credit.
       Other factors, such as the importance of the retailer, length of the relationship, and
proximity to the wholesaler, also play a role in wholesalers’ decision to provide credit and
its terms. Wholesalers are more likely to provide trade credit and provide larger amounts
of trade credit to retailers who account for a greater share of their total sales, and are
therefore more important to the wholesaler’s business. Wholesalers also provide such retailers
credit after fewer cash transactions, but require them to repay within a shorter time. These
findings suggest that although wholesalers provide larger amounts of credit to more important
retailers, they also employ caution in offering terms to such retailers. In addition, wholesalers
are more likely to provide trade credit and offer nonfixed repayment times to retailers with
whom they have a longer relationship. Interestingly, wholesalers were less likely to have
provided these retailers with trade credit the first time they transacted and also provided
them credit after a greater number of cash transactions.11 Whether the wholesaler and
retailer are the same gender does not significantly impact the access to and terms of trade
credit.
       Finally, proximity to the wholesaler matters in access to credit. Wholesalers are more
likely to provide credit at the first transaction and after a lower number of cash transactions
to retailers who are located closer to their shops, relative to retailers that are located farther

  11
    The lending relationships are not stronger for within-community wholesalers and retailers. The correla-
tion between Same Community and Relationship Length is negative and insignificant, further highlighting
that the effect of community membership cannot be explained by existing lending relationships.

                                                    20

away.12 These results can be explained by wholesalers having personal relationships with
retailers whose shops are located close by, as personal relationships in lending have been
documented to increase credit availability (Fisman et al., 2017; Haselmann et al., 2018). We
also find that wholesalers are more likely to experience defaults from proximate retailers.
This higher default frequency is also in line with lending based on personal relationships,
which may lead to worse lending decisions (Haselmann et al., 2018).

3.2. Actions taken in case of default

      We asked wholesalers about the actions they took if retailers defaulted, based on the
161 cases of default in our sample. When asked whether they talked with specified parties
about defaults, no wholesalers reported that they talked to other wholesalers, talked to
their community members, or talked to community members of retailers about the defaults
(Table 3, Panel A). In cases where wholesalers did take some action, 39% of the time they
attempted to persuade the retailer to repay. Other actions include repossessing the goods
(6%), threatening or putting pressure on the retailer (3%), and refusing future credit (3%).
There is also no evidence that wholesalers take legal action against defaulting retailers,
consistent with weak legal systems. Overall, we find that of the 161 cases of default in our
sample, wholesalers took action in only 43% of the cases.
      In Table 3, Panel B, we explore the actions taken by wholesalers in case of default in a
multivariate model. The majority of wholesalers in our default sample do not face multiple
defaults. Therefore, instead of wholesaler fixed effects as in our primary analyses in Ta-
ble 2, we rely on product and wholesaler-community fixed effects to capture time-invariant
industry and community characteristics that may explain wholesalers’ actions in the event
of default. The dependent variables are indicators corresponding to a particular action. In-
terestingly, the results show that wholesalers are more lenient towards retailers from their
community. For instance, the negative and significant coefficient on Same Community in col-

 12
    The correlation between Same Community and Distance (Close) is −0.11397 (p-value of < 0.0001),
indicating that same-community retailers are not located closer to wholesalers’ shops.

                                               21

umn (1) suggests that wholesalers are 23% less likely to use persuasion to deal with defaulting
retailers from within their community relative to defaulting retailers from a different com-
munity. Wholesalers are also less likely to take other actions, including repossessing goods,
threatening or putting pressure, or denying future credit against same-community retailers
who default (columns (2)–(4)). Overall, wholesalers are 29% less likely to take any action
against defaulting retailers from within their community than against defaulting retailers
from outside communities (column (5)).
   To summarize, our findings in Table 2 and Table 3 suggest that, controlling for the other
important determinants of the access to trade credit and its terms, there is preferential lend-
ing within the community in terms of the likelihood and the amount of term credit. Further-
more, our finding that wholesalers are more lenient towards same-community retailers in case
of default might imply that a rational retailer should repay outside-community wholesalers
before same-community wholesalers, resulting in higher within-community defaults. In con-
trast, we find that wholesalers face lower defaults from same-community retailers. Overall,
these findings suggest the presence of a mechanism that incentivizes same-community whole-
salers and retailers to cooperate. We conjecture that a reciprocity mechanism explains such
cooperation, and provides insurance against unexpected shocks to traders’ income. This
reciprocity mechanism combines elements of community dependence and greater within-
community information flows, as we elaborate below.

4. Reciprocity and lending to community members

4.1. Discussion

   A norm of reciprocity implies that individuals who follow this norm are obligated to
behave reciprocally, and deviation results in a reduced ability to rely on community members.
This mechanism is distinct from a pure economic exchange because the currency may not
be monetary and the timing of repayment is undetermined (Portes, 1998). Importantly,
reciprocity could be indirect and extend beyond bilateral agreements, which imply repeated

                                              22

interactions between two individuals, to a more generalized view that extends to nontrading
members of the group (Ferrara, 2003). Such indirect reciprocity has also been demonstrated
in experimental research. For example, in experimental settings of donor–recipient games,
recipients who have helped others in the past get greater transfers from donors, who base
their decision on recipients’ helping behavior (Bolton et al., 2005; Engelmann & Fischbacher,
2009; Seinen & Schram, 2006; Wedekind & Milinski, 2000). These studies highlight the
importance of transmitting information about members’ helping behavior to third parties,
suggesting that indirect reciprocity is closely linked to reputation. That is, the mechanism of
indirect reciprocity channels support towards those who build a reputation for collaboration
(i.e., those who helped others in the past) and at the same time, provides an incentive to
cooperate because noncompliance is associated with credible threats to withdraw cooperation
(Nowak & Sigmund, 2005). Thus, in our setting, a wholesaler who is lenient towards same-
community retailers may receive benefits not only from that particular retailer, but also from
others in the community who are not direct beneficiaries in the current transaction.
       Indirect reciprocity is critical in our setting because credit allocation within communi-
ties serves as an informal insurance mechanism to smooth income shocks over time (Besley
& Coate, 1995; Fafchamps & Lund, 2003; Kinnan & Townsend, 2012; Udry, 1994).13 If
idiosyncratic shocks to income occur, in the absence of social security or other kinds of re-
liable formal safety nets, traders can smooth their income ex post by relying on community
members for aid. Such insurance is not costless as wholesalers, by extending larger amounts
of trade credit to same-community retailers, may need to obtain additional resources to
finance their operations. Similarly, by repaying same-community wholesalers over outside-
community wholesalers, retailers forgo the benefits of the leniency from same-community
wholesalers. We argue that the mechanism of reciprocity explains the cooperative relation-
ship between same-community retailers and wholesalers, such that wholesalers lend more

  13
   The importance of reciprocity in informal economies is further supported by a strong ethic of reciprocity
among the poor as an adaption to poverty (Desmond, 2012; Stack, 1975).

                                                    23

to within-community retailers and are more lenient when these retailers default and that
retailers repay these wholesalers.

4.2. Community dependence, information flows, and incentives to reciprocate

       Indirect reciprocity is difficult to test because there is no single outcome that measures
the benefits traders receive. Therefore, to test for the presence of this mechanism, we take
an approach based on wholesalers’ incentives to reciprocate: those who are more likely to
benefit from the norm of reciprocity are also more likely to follow it. Specifically, we expect
reciprocity to be stronger among wholesalers who are more dependent on their community
for insurance against idiosyncratic shocks to income. We use a number of measures of
wholesalers’ community reliance to assess whether a reciprocity mechanism affects their
lending to members within the community.
       We expect the mechanism of reciprocity to be stronger among wholesalers who do not
have a stable source of income (e.g., rental income, family members with a stable job) and
are therefore more likely to be reliant on their community in times of economic distress.
Also, wholesalers with low education levels, defined as a few years of schooling or no formal
education, have fewer outside options if their business fails. Therefore, we expect reciprocity
and reliance on the community to be higher among such wholesalers. These predictions are
in line with Munshi & Rosenzweig (2016), who show that households with lower income and
those with a higher variation in their income benefit to a greater extent from the community
insurance network. Furthermore, wholesalers who report that they also engage in retail
business are likely to be direct beneficiaries of reciprocity as they not only provide credit as
part of their wholesale business, but also require credit for their retail business.14
       To test for the presence of reciprocity, we re-estimate Equation 1 for the issuance of

  14
     One could argue that wholesalers that engage in retail business could be funding their competitors by
offering credit to retailers. However, such wholesalers have substantially smaller retail businesses relative
to their wholesale operations, which cannot be sustained without providing credit to retailers who typically
cannot pay upfront for large amounts of goods. This suggests that the rewards from providing credit to
retailers exceed the costs.

                                                     24